Reducing a model to only one-two terms

[#1]

Hello all,

I'm currently analyzing a 4 factor half-factorial design. My pareto chart is indicating that only one-two main effects for given responses are significant. Would it be acceptable to remove the remaining insignificant main effect and interaction terms from the model?

I've trialed this, by reducing the model as described above. The R2 and Adjusted R2 are close (0.68 vs 0.63), so not fantastic. However, I don't think I should be adding in the insignificant terms and risk overfitting the model. Could anybody advise? 

Thank you

[#2]

Without seeing the other diagnostic information or the actual output from your experiment, it is difficult to give a specific set of advice.  That being said, it would typically be appropriate to remove insignificant terms from the model.  Do the results make sense?  Are the factors practically significant? How do the results compare to your predictions? How much did the R-square adj change when you removed the terms?  What do the residuals look like?  Leverage plots? What happened to the RMSE?